AWS Design Patterns at Anki

Transcript

1. AWS Design Patterns at Anki Advanced AWS Meetup 1/21/2014 Ben

   Whaley | @iAmTheWhaley | ben@anki.com

2. Work for Anki! Anki needs beta testers!

3. • Anki and AWS • Application and Systems design •

   Deployment and Provisioning • Demo • Partnerships Overview
4. None

5. Services, analytics and infrastructure • Provides web & analytics services

   for Anki • Example services: ◦ Gameplay diagnostics ◦ Game replays ◦ Customer service platform ◦ Future Anki DRIVE enhancements

6. Application design

7. • Application processes are stateless Application design

8. • Application processes are stateless • Loose coupling among services

   Application design

9. • Application processes are stateless • Loose coupling among services

   • Configuration is in the environment Application design

10. • Application processes are stateless • Loose coupling among services

    • Configuration is in the environment • Built for continuous deployment Application design

11. Systems design

12. • n+2 redundancy Systems design

13. Systems design • n+2 redundancy • Compute resources are ephemeral

14. Systems design • n+2 redundancy • Compute resources are ephemeral

    • Managed services preferred

15. Systems design • n+2 redundancy • Compute resources are ephemeral

    • Managed services preferred • Infrastructure as code

16. Systems design • n+2 redundancy • Compute resources are ephemeral

    • Managed services preferred • Infrastructure as code • Single purpose VPCs

17. Systems design • n+2 redundancy • Compute resources are ephemeral

    • Managed services preferred • Infrastructure as code • Single purpose VPCs • Contract between backing resources and applications

18. Backing resources Public Web Services ELBs Infrastructure Services Bastion, NAT,

    CM, Metrics Private Web Services Persistence VPC External AWS Services S3, SQS, SNS Inbound requests from clients

19. Deployment and Provisioning

20. Goal: Continuous deployment Source Repo commit GitHub Web Hook Deploy

    Worker Old Release Remains Running Integration Tests Deploy to Integration Deploy to Production Integration tests pass SQS New Release Live Integration tests fail Notify Check for commits Commit found DNS Change

21. Deployment and Provisioning • New code, new instances

22. Deployment and Provisioning • New code, new instances • Instance

    store-backed AMIs built with Packer

23. • New code, new instances • Instance store-backed AMIs built

    with Packer • AMIs ship with base Docker containers Deployment and Provisioning

24. Deployment and Provisioning • New code, new instances • Instance

    store-backed AMIs built with Packer • AMIs ship with base Docker containers • Instances provisioned with Ansible

25. Userdata: 1. Sets configuration variables 2. Downloads ansible playbooks 3.

    Runs ansible 3-step provisioning pipeline Deploy tool: 1. Loads configuration from S3 2. Sets EC2 userdata bash script 3. Launches instances Ansible: 1. Configures supporting software 2. Downloads application code 3. Builds & runs Docker containers 4. Sends notifications

26. • Base container, service container, release container • Local dev

    == Production • Exceptional community • Challenges ◦ Rapidly changing ◦ Logging ◦ Debugging

27. • Uses troposphere library to create CloudFormation templates • Reusable

    resource templates • Saves app configuration as json files in S3 • Creates CF stacks & kicks off instance provisioning • Supports multiple deployments & releases Custom deployment tool

28. • IAM roles and policies created via CloudFormation • Intra-VPC

    traffic authorized by SG ID • Precise subnetting allows for future network ACLs • … some credentials still laying around in plain text On security

29. VPC Persistence Persistence Persistence RDS Multi-AZ Persistence Persistence Private Web

    Services Queue Processor Query Tool Persistence Persistence Public Web Services ELB Persistence Persistence Infrastructure Services Bastion NAT SQS Inbound Requests Flickr Demo App

30. Deployment Demo

31. Tooling & design limitations • Blue/Green deployment & ELB pre-warming

    • “Static” infrastructure templates • Higher latency for requests between VPCs • Ephemeral instance tradeoffs ◦ Slower to boot than EBS ◦ Slower to build AMIs • Incomplete functionality (for now)
32. None

33. VPC Persistence Persistence Persistence Persistence Persistence Private Web Services Persistence

    Persistence Public Web Services Persistence Persistence Collector Log collection and analytics

34. Credits • The 12 Factor App - http://12factor.net • Werner

    Vogels 2012 re:Invent Keynote • Resty - https://github.com/micha/resty • Troposphere Python cloudformation library ◦ https://github.com/cloudtools/troposphere/ • Continuous deployment at Quora ◦ http://engineering.quora.com/Continuous- Deployment-at-Quora